Author contributions: P.R.-O.: collection of data and data analysis and interpretation; J.L.O.: collection of data and provision of study material; B.A., A.E.-A., M.T.B., J.A.M., L.G., L.N., S.D., and A.E.-O.: collection of data; J.M. and A.V.-B.: collection of data and provision of study patients; M.L., A.P., J.A.M.-C., and M.C.: data analysis and interpretation; V.S.: assembly of data; V.S.-M.: conception and design and data analysis and interpretation; J.L.F.-L.: conception and design, data analysis and interpretation, and manuscript writing.
Disclosure of potential conflicts of interest is found at the end of this article.
first published online in STEM CELLS EXPRESS February 8, 2013.
Glioblastoma (GBM) is associated with infiltration of peritumoral (PT) parenchyma by isolated tumor cells that leads to tumor regrowth. Recently, GBM stem-like or initiating cells (GICs) have been identified in the PT area, but whether these GICs have enhanced migratory and invasive capabilities compared with GICs from the tumor mass (TM) is presently unknown. We isolated GICs from the infiltrated PT tissue and the TM of three patients and found that PT cells have an advantage over TM cells in two-dimensional and three-dimensional migration and invasion assays. Interestingly, PT cells display a high plasticity in protrusion formation and cell shape and their migration is insensitive to substrate stiffness, which represent advantages to infiltrate microenvironments of different rigidity. Furthermore, mouse and chicken embryo xenografts revealed that only PT cells showed a dispersed distribution pattern, closely associated to blood vessels. Consistent with cellular plasticity, simultaneous Rac and RhoA activation are required for the enhanced invasive capacity of PT cells. Moreover, Rho GTPase signaling modulators αVβ3 and p27 play key roles in GIC invasiveness. Of note, p27 is upregulated in TM cells and inhibits RhoA activity. Gene silencing of p27 increased the invasive capacity of TM GICs. Additionally, β3 integrin is upregulated in PT cells. Blockade of dimeric integrin αVβ3, a Rac activator, reduced the invasive capacity of PT GICs in vitro and abrogated the spreading of PT cells into chicken embryos. Thus, our results describe the invasive features acquired by a unique subpopulation of GICs that infiltrate neighboring tissue. STEM Cells2013;31:1075–1085
Invasion of glioblastoma (GBM) into healthy tissue restricts therapeutic intervention and surgical resection. Even after macroscopically complete resection of the tumor, local invasiveness eventually leads to regrowth of a recurrent tumor, which is not significantly altered by radiation or chemotherapy . GBM is frequently associated with significant involvement of peritumoral (PT) parenchyma infiltrated by isolated tumor cells [2–4]. Moreover, analyses of GBM-affected brains have demonstrated the presence of tumor cells deep in the contralateral hemisphere . Current therapeutic regimens do not adequately address the disseminated disease burden associated with infiltrative GBMs, and there is, therefore, an urgent need to develop novel treatment approaches to specifically target the invasive capacity of these tumors. Work in GBM and other cancers supports the presence of cancer-initiating or stem cells which possess innate resistance mechanisms against treatment-induced cell death, allowing them to survive and initiate tumor recurrence . However, the cellular and molecular determinants of cancer-initiating cell migration and invasion in GBM are not known, and this may be a key issue to design therapeutic strategies against disseminated disease. In line with this, it has been recently shown that GBM cells with stem-like features can be obtained from the resection margin of GBM tumors . Thus, efforts should be focused to unravel the cellular mechanisms contributing to provide GBM-initiating cells (GICs) with the capacity to infiltrate the neighboring tissue. A huge variety of intracellular signaling molecules have been implicated in cell migration and invasion. However, Rho-family GTPases, Rac, Cdc42, and RhoA, are considered to play a pivotal role in regulating the biochemical pathways most relevant to these processes . Recent studies, using in vivo fluorescence resonance energy transfer imaging, have revealed two different modes of invasion at perivascular and intraparenchymal regions of a rat GBM cell line . In this model, GBM cells penetrating the brain parenchyma showed higher Rac1 and Cdc42 activities and lower RhoA activity than those advancing in the perivascular regions. Nevertheless, it has also been described that functional inhibition of Rac1 does not suppress GBM migration, and dominant positive forms of Rac1 block motility of GBM cells . The interpretation of the existing data acquires even more complexity when we consider that those features controlled by Rho GTPases, including protrusion formation and cell motility, depend on properties of the microenvironment, such as matrix composition, rigidity, and the use of two-dimensional (2D) or three-dimensional (3D) matrices [11, 12]. The dynamics of Rho GTPase activities are tightly regulated in order to achieve their specific subcellular localization . Integrin engagement stimulates the activity of Rho family GTPases, promoting actin assembly . Moreover, αVβ3 integrin promotes Rac1 activation in glioma cells  and this integrin has been shown to be involved in the sensing of matrix rigidity at the leading edge during cell spreading . Additionally, p27 is able to bind to RhoA, thereby inhibiting RhoA activation by interfering with the interaction between this GTPase and guanine nucleotide exchange factors . This activity of p27 is independent of its cell-cycle regulatory functions and depends on its cytosolic localization. In this study, we found that PT GICs have an invasive advantage over GICs from the tumor mass (TM) and this invasion capacity is sustained by upregulation of αVβ3 integrin and downregulation of p27.
MATERIALS AND METHODS
Patients and Surgical Specimens
Patients diagnosed of GBM multiforme were selected for the study after informed written consent had been given, as approved by the Research Ethics Board at the Valdecilla University Hospital. Craniotomy was tailored according to the exact extension of the tumor. A neuronavigation system was used to guide tissue sampling. Normal white matter, enhancing TM, and the PT region were defined on the basis of imaging features. Normal white matter was defined as a normal-appearing area that contained no enhancement and showed normal signal intensity on T2-weighted imaging. The enhancing TM contained a solid portion, preferably with avoidance of necrotic components. PT area was considered a region clearly outside the well-defined enhancing solid portion that contained absolutely no enhancement and had high signal intensity on T2 imaging. Tissue from the PT region was obtained first, thus avoiding sample contamination from the TM.
Cell cultures from tissue specimens were established as previously described . Briefly, tissue was subjected to enzymatic digestion, and then tumor cells were resuspended in serum-free Dulbecco's modified Eagle's medium (DMEM)/F-12 medium (Invitrogen, Carlsbad, CA, www.invitrogen.com) containing human recombinant epidermal growth factor (20 ng/ml; Sigma-Aldrich, St. Louis, MO, www.sigmaaldrich.com), basic fibroblast growth factor (20 ng/ml; Sigma-Aldrich), B-27 (20 μl per ml of medium; Invitrogen), and heparin (2 μg/ml), and plated at a density of 3 × 106 live cells per 60-mm plate. Primary neurospheres were detected within the first 2 weeks of culture and subsequently dissociated every 3–4 days to facilitate cell growth. All cell cultures underwent clonal analysis at clonal density of 100 cells per milliliter to avoid clumping. The number of cells in each culture was determined every 48 hours up to 10 days and the mean population doubling time calculated as described . To promote differentiation, neurospheres were cultured in the same medium but in the presence of 10% fetal calf serum (FCS) for 4 days. When indicated, cells were incubated with 5 μM H1152 (Tocris Bioscience, Bristol, U.K., http://www.tocris.com), 30 μg/ml of blocking antibodies against αVβ3 integrin (LM609; Millipore, Billerica, MA, http://www.millipore.com), or irrelevant control antibody. Human umbilical vein endothelial cells (HUVECs) (kindly provided by Dr. Samuel Cos, University of Cantabria, Spain) were grown to confluent monolayer in endothelial-cell culture medium as previously described . HUVECs were cocultured with GICs previously stained with 2 μM carboxyfluorescein succinimidyl ester (CFSE; Life Technologies, Grand Island, NY, www.lifetechnologies.com). Tumor cells in suspension and cells recovered after washing with 5 mM EDTA were pooled (Fraction F1). Transmigrating and/or strongly adherent tumor cells were released from the monolayer following incubation with trypsin (Fraction F2). Absolute tumor cell counts were determined by flow cytometry, through gating CFSE-positive cells, and counting beads (CountBright; Life Technologies) according to the manufacturer.
Cells were nonenzymatically removed with Cell Dissociation Buffer (Life Technologies) before reseeding onto 12-well plates coated with 1.8 mg/ml Collagen type I. After 15 minutes, nonadherent cells were removed, and adherent cells were fixed using 4% paraformaldehyde (PFA) and counted in four ×20 fields per culture.
Total RNA from TM and PT GICs was extracted using the RNeasy mini kit (Qiagen, Valencia, CA, http://www.qiagen.com) and microarray gene expression analysis was performed with the Human Genome U133 Plus 2.0 array (Affymetrix, Santa Clara, CA, www.affymetrix.com). Both background correction and normalization were done using Robust Multichip Average algorithm. The selection of those genes differentially expressed between TM and PT cells was performed using a criteria based on the fold-change value. Probe sets were selected as significant using a logFC cut-off of 1.5. The raw data have been deposited in a MIAME compliant database (GEO accession number, GSE37985).
To assess the expression of individual genes, a cDNA was generated and amplified using primers for human CD133, Sox2, Nanog, Tuj1, glial fibrillary acidic protein (GFAP), β-Actin , p27 (5′TGGAGAAGCACTGCAGAGAC3′) and (5′GCGTGTCC TCAGAGTTAGCC3′), and integrin β3 (5′GCAGCTGTGGGGAC TGCCTG3′) and (5′TGGGCAGGTGGGGCACTTCT3′). Quantitative real-time PCR was performed in a 7000 Sequence Detection System (Life Technologies) as described . For the differentiation markers, data were represented as fold change in relative expression levels as compared to undifferentiated (GIC) cells. At least three independent experiments for each quantitative PCR were performed.
Western Blot and Pulldown Assays
Cell extracts were separated on a 15% polyacrylamide gel and transferred to Polyvinylidene Difluoride (PVDF). Blots were blocked with 5% bovine serum albumin (BSA) and incubated with mouse monoclonal antibodies against Rac1 (Millipore), RhoA, αTubulin, lamin A/C (Santa Cruz Biotechnology, Santa Cruz, CA, www.scbt.com), or rabbit anti-p27 and anti-Cdc42 antibodies (Santa Cruz), followed by incubation with goat anti-rabbit or anti-mouse antibodies conjugated to alkaline phosphatase (Sigma-Aldrich). Bound antibody was detected by a chemiluminescence system (Life Technologies). Protein extracts from cytoplasmic and nuclear fractions were obtained as described . Protein band quantification was carried out using ImageJ software. Glutathione S-transferase-conjugated Rhotekin and PAK-CRIB were expressed from pGEX-Rhotekin and pGEX-CRIB, and assays were performed as described .
Single nucleotide polymorphism (SNP) Microarray Analyses
DNA copy number changes were evaluated using Affymetrix GeneChip 250-NspI/StyI SNP microarrays. Hybridization, washing, and signal detection were performed in a GeneChip Fluidics Station 400 and GeneChip scanner 3000 according to the manufacturer's protocols (Affymetrix). To identify DNA copy number changes, bioinformatic analyses were performed using Affymetrix Genotyping Console 3.0.2 and Partek 6.5 software (Partek Incorporated, St. Louis, MO, http://www.partek.com), as reported previously .
Gene Silencing and Transfections
Cells (2.5 × 105) were transfected with 20 nM oligonucleotide pools specific for Rac1, Rac3, or β3 (SmartPools from Dharmacon, Thermo Fisher Scientific, Waltham, MA, www.thermoscientific.com), with Optimem-I and Lipofectamine 2000 (Life Technologies). Cells were plated in collagen after 2 days of transfection and their invasion capacity was analyzed 24 hours later as described below. When indicated, cells were transfected with p27 shRNA-containing pSM2C plasmid  or pEGFP-QL RhoA  using Amaxa nucleofector and reagents (Lonza, Basel, Switzerland, http://www.lonza.com).
Cells were assayed for the expression of Nestin, GFAP, Tuj1, and Sox2 by fluorescence or confocal microscopy as described previously . Briefly, cells were fixed in 3.7% formaldehyde and incubated overnight with rabbit anti-GFAP (DAKO, Glostrup, Denmark, www.dako.com), or mouse anti-Nestin, anti-Sox2 (both from R&D Systems, Minneapolis, MN, http://www.rndsystems. com), and anti-Tuj1 (Sigma-Aldrich). Then, Texas red-conjugated or FITC-conjugated goat anti-rabbit or anti-mouse secondary antibodies (Jackson ImmunoResearch, Cambridgeshire, U.K., http://www.jacksonimmuno.com) were used for fluorescence detection.
Cell Culture on Thick Deformable Matrices
Bovine skin collagen I (PureCol 5005-B, Nutacon BV, Leimuiden, The Netherlands, www.nutacon.nl) was prepared at 1.8 mg/ml in serum- and phenol-free DMEM. Following polymerization at 37°C, cells were seeded on top of matrix in medium, allowed to adhere for 2 hours, and imaged for the indicated time intervals.
Cell Motility Assays
Polyacrylamide matrices of different rigidities were prepared as previously described . All hydrogels were covalently functionalized with laminin (Sigma-Aldrich) at a nominal surface density of 1.4 μg/cm2. A rheometer (Anton Paar, Osterreich, Austria, http://www.anton-paar.com) was used to measure the macroscopic elastic shear modulus of each gel. Cell motility was evaluated by tracking the movement of individual cells at 20-minute intervals over a 24-hour time course in a Ti-Eclipse Microscope. The tracking module of the NIS Elements (Nikon, Melville, NY, http://www.nikon.com) software was used to track automatically the centroid of each cell throughout the time sequence.
Cells were suspended in serum-free matrix composed of 2.3 mg/ml growth factor-reduced Matrigel (BD Biosciences) or hyaluronic acid (HyStem; Sigma-Aldrich) and 1.1 mg/ml collagen I or collagen I only at 2.2 mg/ml to 10,000 cells/100 μl. Aliquots were dispensed into 96-well ViewPlates (PerkinElmer, Waltham, MA, http://www.perkinelmer.com) and processed as described previously . Confocal Z slices were collected at 50 μm and 3 μm (bottom of well) with Ti-Eclipse microscope (Nikon). To obtain images of the collagen matrix, confocal reflectance microscopy was used. An argon laser at 488 nm was the excitation source, and the reflected light (collagen fibrils) was detected. Three images per gel were collected.
For conventional electron microscopy, vibratome sections (200-μm thick) containing tumor tissue were postfixed first in 3% glutaraldehyde and then in 2% osmium tetroxide, dehydrated in acetone, and embedded in Araldite. Semithin sections (1 μm) were stained with toluidine blue for light microscopy examination. Ultrathin sections stained with uranyl acetate and lead citrate were examined with a Philips EM-208 electron microscope.
Tumor Xenografts and Imaging in a Mouse Model
The mouse model of human GBM was established in the Animal Core Facilities of the Center for Applied Medical Research (University of Navarra) after approval by the Institutional Animal Ethics Committee. Briefly, 1 × 106 GICs stably transfected with luciferase- and CherryFP-containing plasmids  were injected into the brain of anesthetized 6–8-week-old female BALB/cA-RAG2(−/−)-IL2γc(−/−) mice as described previously . To monitor intracranial tumor growth, noninvasive bioluminescence imaging was initiated 10 days after tumor cell implantation and repeated once a week using the IVIS Spectrum system (Advanced Molecular Vision, Lincolnshire, U.K., http://www.amv-europe.com).
Tumor Xenografts in a Chicken Model
Fertilized chicken eggs of 5.5 days of incubation were windowed and labeled cells were microinjected into the embryos limb bud, at the distal tip of the third interdigital space. After injection, eggs were sealed and incubated for further 24 hours. Embryos were then fixed in 4% PFA, limbs dissected free, and monitored by confocal microscopy. Limbs were optically sectioned longitudinally along dorsal–ventral planes at 15 μm intervals. For stacks digitalization and image processing, we used the LSM 5 Image Examiner software. Images shown in this work are the integration of all the Z-stacks taken to cover the whole limb.
All statistics were calculated with the SPSS statistical package (version 13.0). Quantitative analyses of cell migration, adhesion, and morphology included at least 100 individual cells unless otherwise indicated. Data are presented as mean ± SD. Differences between groups were tested for statistical significance using the unpaired two-tailed Student's t test. The significance level was set at p < .05.
Identification and Characterization of GICs Derived from the TM and the PT Area
Magnetic resonance imaging of GBMs (Fig. 1A) showed a TM surrounded by a PT region, which corresponds to parenchyma infiltrated by isolated tumor cells [2, 3]. Most TM is removed during surgery, but the infiltrative disease associated with the PT tissue remains. We identified stem-like TM and PT cell populations using the neurosphere assay, a culture system extensively used in previous studies [26–28]. Here, we will refer to these populations as GICs, defined by their ability to form tumors following xenotransplantation into immunodeficient mice , as shown below. We established paired cultures of TM and PT GICs from three patients diagnosed of GBM multiforme (GIC1, GIC2, and GIC3). Clonal analysis of all GIC neurosphere cultures demonstrated that they were capable of producing new neurospheres (not shown). Then, we determined the stem-like properties of both cell populations. TM and PT GICs formed neurospheres in culture that expressed neural stem cell markers Nestin and Sox2 in more than 80% of cells (Fig. 1B; supporting information Fig. S1), and were able to undergo differentiation toward astrocytic and neuronal lineages as shown by a reduction in the levels of stem cell-related markers Sox2, Nanog, and CD133, and an increase in the expression of the differentiation markers GFAP and Tuj1 (Fig. 1C; supporting information Fig. S1). We analyzed TM and PT cell populations for the presence of copy number variations using microarray-based comparative genomic hybridization and identified virtually the same genomic changes in both cell populations that mostly involved chromosomes 1, 7, 9, 19, and 20 (Fig. 1D). Copy number variations of these chromosomes have already been described in GBM neurosphere lines . Furthermore, both GIC cultures retained the tumor formation capacity as determined by bioluminescence imaging after injecting cells, stably expressing luciferase, into the brain of immunodeficient mice (Fig. 1E). Of note, TM cells generated bigger tumors than PT cells. Consistently, TM cells displayed reduced doubling time compared with PT GICs (Fig. 1F). These data were also consistent with a higher proportion of cells in S and G2/M, a higher diameter of the neurospheres over time and an increase in glucose consumption to meet energetic and biosynthetic demands (data not shown). Thus, the TM cell subpopulation displays a proliferation capacity higher than that observed in PT cells. An inverse correlation between proliferation and migration or infiltration capacity of tumor cells has been already described [31, 32]. A low proliferation rate could be important for tumor cells while they infiltrate surrounding tissue.
As a first attempt to demonstrate differences between both GIC populations at the molecular level, we searched for genes that performed best at distinguishing between TM and PT cells in a gene expression microarray (Fig. 1G). Linear models of microarray analysis identified 64 genes with significant differences (p < .01), associated with a number of biological functions, including cell adhesion, migration, cell cycle, and immune or inflammatory responses. Upregulated genes in PT GICs included Neurexin 1, Cadherin 20 (CDH20), dipeptidyl peptidase 4, Snail homolog 2 (SNAI2), Nanog homeobox (NANOG), disrupted in schizophrenia 1 (DISC1), and Protocadherin 19 (PCDH19). Conversely, a number of genes including Palladin (PALLD), Nuak1 kinase (NUAK1), Neuropilin 1, phosphodiesterase 8B, Toll-like receptor 4 (TLR4), Amyloid precursor-like protein 1, and Cyclin-dependent kinase inhibitor 1B (CDKN1B) were found consistently downregulated in PT GICs. From this point on, most of the results shown were obtained using GIC1 cultures, unless otherwise indicated.
Cell Motility and Invasiveness Are Increased in PT Cells
The aggressiveness of GBM is partly derived from infiltration of tumor cells into the surrounding brain parenchyma, which is less rigid than the TM . Microenvironment stiffness has a strong influence on stem cell behavior  and regulates glioma cell motility . To address the response of both GIC populations to changes in the mechanical properties of the microenvironment, we first studied migration of cells cultured on matrices of variable stiffness, from 0.7 kPa, which resembles the rigidity of normal brain tissue, to 10 kPa. We observed that migration speeds fell with decreasing substrate rigidity in TM cells. However, migration of PT cells does not depend on changes in stiffness (Fig. 2A, 2B), suggesting that they have more efficient mechanotransduction mechanisms, which may facilitate infiltration into less rigid neighboring tissue. Moreover, PT cells display a high plasticity in protrusion formation and migrate in a fashion in which the cell thins and extends as it advances (Fig. 2C; supporting information Videos S1 and S2). We also explored the invasive behavior of cultured neurospheres within collagen matrices (supporting information Fig. S2A). Both collective and individual invasion were enhanced in PT cells (supporting information Fig. S2B), which formed readily detectable collective strands (supporting information Fig. S2C). PT cells were also more invasive when neurospheres were dissociated into individual cells and allowed to migrate through collagen-based matrices (Fig. 3A). The increase in invasion displayed by PT cells was also detected when collagen was combined with either Matrigel or hyaluronic acid. However, the strongest advantage was observed in matrices containing collagen only (Fig. 3B), a matrix that favors tumor cell plasticity . Morphological analysis in 2D (Fig. 3C) and 3D (Fig. 3D) collagen matrices revealed long actin-filled protrusions in PT cells. Moreover, the average length of protrusions in TM cells was about half of the length seen in PT cells (Fig. 3E). PT cells also showed higher adhesion (mean of 2.5-fold) when plated on a thick collagen matrix (Fig. 3F), which is likely due to increased integrin-mediated signaling. In addition, consistent with our previous data on hydrogel substrates, the migration speed of PT cells on a collagen matrix was higher than that of TM cells (Fig. 3G).
Mouse and Chicken Embryo Xenografts Reveal the Invasive Nature of PT Cells
TM and PT GICs stably expressing cherry fluorescent protein were injected intracraneally into immunodeficient mice. Subsequent histological analysis revealed that TM cells formed small TMs, whereas PT cells showed a dispersed distribution pattern, with single or clustered cells infiltrating the cortex and periventricular region within the mouse brain (Fig. 4A). Electron microscopy of ultrathin sections revealed that nuclei of PT cells were markedly enlarged and irregularly shaped, with frequent bizarre forms, a feature associated with malignancy  and tumor cell invasion , as opposed to TM cells, with less irregular nuclei (Fig. 4A). PT cells were also seen in proximity to blood vessels by electron microscopy, which was not observed in TM cells (supporting information Fig. S3A). The proportion of PT cells located within 2 μm from the blood vessel wall was approximately 37%, whereas no TM cells were observed considering the same distance limit. Further evidence of the interaction between PT and endothelial cells came from the colocalization of fluorescent signals corresponding to lectin-staining endothelial cells and CherryFP-expressing PT cells (supporting information Fig. S3B). Moreover, when GICs were cultured on confluent monolayers of HUVEC endothelial cells, PT GICs showed stronger adherence than TM GICs (approximately 30% more strongly adherent cells) as determined by the stringent conditions required to detach tumor cells (supporting information Fig. S3C--S3E). The in vivo data were further confirmed in a chicken embryo xenograft model obtained by microinjecting DiI-labeled GICs in the embryos limb bud (Fig. 4B). Consistent with the mouse model, PT but not TM cells invaded the embryo tissues, spreading through the interdigital space. Interestingly, PT GICs were observed along the axial vessel, further confirming the close interaction between PT cells and vascular endothelium. Overall, the animal models reveal an advantage of PT cells to invade surrounding tissue and suggest that blood vessels might help outline their migration pathway.
Rac and RhoA Signaling Mediate the Invasive Capacity of PT Cells
Rho GTPases are key regulators of cell migration and invasion due to their roles controlling the actin cytoskeleton. We first determined their activation in TM and PT cells by measuring the level of GTP-bound Rho proteins. Figure 5A, 5B, shows increased Rac1 and RhoA activation in PT cells, whereas Cdc42 displayed a similar level of activation in both cell populations. Taking into account that rounded or amoeboid movement is associated with high levels of Rho-ROCK signaling, whereas elongated movement is associated with Rac activation [21, 37], this result is consistent with a high plasticity of PT cells. To further study the contribution of these GTPases to the invasion capacity of GICs, we silenced Rac1 in PT cells by siRNA (Fig. 5C, 5D), and analyzed the invasive potential of tumor cells. Rac1-depleted PT cells displayed a modest reduction in invasive capacity compared with control cells (Fig. 5E), which was further decreased when Rac1 knockdown cells were incubated with H1152, an inhibitor that blocks ROCK signaling downstream of RhoA. It has been shown that both Rac1 and Rac3 are relevant for the invasive behavior of cancer cell lines, including GBM cells [38, 39]. Simultaneous silencing of both GTPases (Fig. 5F) did not promote a reduction of cell invasion in comparison with Rac1-depleted cells. However, treatment of double Rac1/Rac3 knockdown cells with H1152 achieved the strongest inhibition of invasion compared to any of the other conditions (Fig. 5E). Overall, these data demonstrate that simultaneous Rac and RhoA activation are required for efficient invasiveness of PT cells.
Upregulation of αVβ3 Integrin and Downregulation of p27 Trigger the Invasive Capacity of PT Cells
Although microarray analysis did not detect consistent differences in integrin expression between TM and PT cells in all three samples, quantitative RT-PCR revealed that β3 integrin was upregulated in PT cells (Fig. 6A). Integrin αVβ3 heterodimer initiates mechanotransduction . Consistently, β3 silencing significantly reduced the invasive capacity of PT cells (Fig. 6B, 6C). αVβ3 has been shown to trigger Rac1 activation in glioma cells . Thus, we incubated PT cells in the presence of αVβ3 blocking antibodies and observed a significant reduction of cell invasion in collagen matrices, which was further reduced when PT cells were cotreated with specific antibodies and the ROCK inhibitor H1152 (Fig. 6D). Invasion of tumor cells in type I collagen matrix is associated with proteolytic denaturation of collagen, which exposes the αVβ3 recognition site amino acids Arginine-Glycine-Aspartic acid (RGD) [41, 42]. An indirect way to prove the exposure of this cryptic motif is to visualize the collagen fibers around cells by reflection microscopy . A dark spherical region around the PT cell body suggests that these cells were able to degrade the immediate surrounding matrix (Fig. 6E). In addition to the in vitro invasion model, we observed that the spreading of PT cells into the embryos limb was clearly reduced when cells were pretreated with blocking anti-αVβ3 antibodies (Fig. 6F).
Since mutations in Rho proteins have not been found in GBM, deregulation of RhoA could occur at the level of expression or activation of its regulators. Among these, cytoplasmic p27 is able to inhibit RhoA activation [16, 44]. Microarray analysis showed that p27 (CDKN1B) was downregulated in PT cells (Fig. 1E), which was further confirmed at the protein level (from 1.3- to 3.8-fold reduction) (Fig. 7A). Moreover, both TM and PT cells accumulate most of the p27 protein in the cytoplasm (Fig. 7B), where is inactive as a growth inhibitor. In order to determine that p27 was inhibiting RhoA activity in GICs, we silenced p27 using shRNA in TM cells (Fig. 7C) and analyzed the subcellular localization of RhoA on the basis that activated RhoA is localized to the plasma membrane . Immunofluorescence with specific antibodies revealed that RhoA was mainly detected on the surface of silenced cells (Fig. 7D). Then, we showed that downregulation of p27 increased the protrusion length (from 27.7 ± 7.3 μm in sh control-transfected cells to 58.6 ± 13 μm in shp27-transfected cells, mean ± SD, p < .0001) (Fig. 7E) and the invasive capacity (Fig. 7F) of TM cells, which is in line with previous work describing that modest or localized RhoA activity may be necessary lo locally activate ROCK and promote persistent membrane protrusions . Consistently, downregulation of p27 also increased the capacity of TM cells to infiltrate surrounding tissue in the embryos limb (Fig. 7G). Thus, a reduction in the levels of p27 could partially revert the low invasive capacity of TM cells, mimicking the PT cell phenotype.
In this work, we have studied whether GICs infiltrating the brain parenchyma represent a unique subpopulation of cells that have acquired invasive capabilities. We observed that the migration capacity of PT cells was not significantly modified by the matrix rigidity, suggesting that these cells may be insensitive to substrate stiffness or have more efficient mechanotransduction mechanisms. By contrast, the migration behavior of TM GICs is consistent with previous work showing that glioma cell lines fail to productively migrate when the extracellular matrix rigidity is lowered to values comparable with normal brain tissue . If we consider the anatomic variation in stiffness within the brain , and that GBM tumors are stiffer than the surrounding parenchyma , we can infer that PT GICs are particularly well suited to infiltrate the PT brain tissue. The capacity of PT cells to invade surrounding tissue was determined using two different xenograft models, mouse brain and chicken embryo. Microscopic analysis of tissue sections revealed that PT cells were in proximity to blood vessels, which might be guiding PT cell migration. In line with this, it has been described that dissemination of glioma cells follows the vascular basement membranes . Furthermore, in our 3D invasion system, the biggest difference in invasion between PT and TM GICs was observed when cells were cultured in collagen matrices. Consistently, the spatial distribution of collagen types I and IV is restricted to the exterior of blood vessels in the brain parenchyma .
We found that Rac1, Rac3, and RhoA GTPases were crucial to promote invasion of PT cells through 3D matrices. On the contrary, a recent model of glioma cell lines revealed that parenchyma-invading cells showed higher Rac1 and lower RhoA activity than those advancing in the perivascular region . It is conceivable that PT GICs are a unique population of stem-like cells selected for its capabilities to activate both Rac and Rho pathways, increasing its plasticity in cell shape and protrusion formation and thus providing cells with a robust mechanotransduction machinery to invade the brain parenchyma.
We have also shown that β3 integrin is upregulated and p27 is downregulated in PT GICs. Rho GTPases can be activated downstream of integrins , and blockade of αVβ3 integrin in different cell types inhibits cell migration  and Rac-dependent invasion . We showed that αVβ3 blocking antibodies reduced the invasion capacity of PT cells in collagen matrices. Invasion of tumor cells in type I collagen matrix is associated with proteolytic denaturation of collagen, which exposes the αVβ3 recognition site RGD [41, 42]. Importantly, αVβ3 blocking strategies are being studied in clinical trials for the treatment of GBM . Conversely, cytoplasmic p27 has been shown to block RhoA signaling both in neuronal progenitors and tumor cells [54, 55]. Moreover, overexpression of p27 in GBM cell lines inhibits cell motility . Interestingly, we found higher levels of cytoplasmic p27 in TM cells and that downregulation of p27 in TM GICs resulted in a significant increase in 3D invasion. Thus, the invasive phenotype of both GIC subpopulations can be partially reverted by modifying the expression or activity of Rho GTPase modulators αVβ3 and p27.
We have also identified a number of genes that may serve to distinguish PT from TM GICs. These markers include genes involved in cell adhesion (CDH20 and PCDH19), migration (SNAI2, NANOG, USP6, and DISC1), and immune or inflammatory responses (TLR4) [57–61], opening new lines of research to deepen our understanding of the molecular features that promote GBM dissemination and progression. A particularly interesting observation is the upregulated expression of HEG1 in PT cells. HEG1 expression is restricted to endothelial cells and it has been shown that HEG1 signaling through KRIT1 is required for blood vessel formation . Notably, GICs are able to transdifferentiate into vascular endothelial cells . It would be of great interest to study whether the increased levels of HEG1 in PT cells and the enhance avidity of these GICs for blood vessels described here are associated with a higher capacity of PT cells to differentiate into endothelial cells.
Using different in vitro and in vivo models, we demonstrate that GICs isolated from the PT area are more capable of migrating and infiltrating the neighboring tissue than GICs from the bulk of the tumor. Moreover, αVβ3 integrin and low levels of cytoplasmic p27 and their downstream effector proteins Rac and RhoA GTPases are needed for PT cells to acquire an invasive advantage. These findings identify a subpopulation of GICs particularly well suited to invade surrounding tissue and describe the signaling pathways that should be the targets of novel therapeutic strategies against GBM dissemination and recurrence.
This work was supported by the Instituto de Salud Carlos III (ISCIII), Spanish Ministry of Science and Innovation Grants PI10/02002, BFU2011-23983, and programs Red Temática de Investigación Cooperativa en Cáncer (RTICC) (Grants RD06/0020/0074, RD06/0020/0041, RD06/0020/0088, and RD06/0020/0046), Centro de Investigación Biomédica en Red sobre Enfermedades Neurodegenerativas (CIBERNED) (Grant CB06/05/0037), and from the Instituto de Formacion e Investigacion Marques de Valdecilla (IFIMAV) Grant API2011-04. Victoria Sanz-Moreno is a Cancer Research U.K. (CRUK) Career Development Fellow. Both V.S.M. and J.L.P. were supported by Grant C33043/A12065.
DISCLOSURE OF POTENTIAL CONFLICTS OF INTEREST
The authors indicate no potential conflicts of interest.